There is growing pressure on agriscience firms to create safer, more efficient, and ecologically friendly goods on a large scale without sacrificing quality or speed.
According to a 2011 baseline, worldwide crop calorie production would need to rise by 47% in 2050 to feed 9.75 billion people under medium population growth, according to a USDA Economic Research Service research. To feed 10.8 billion people under strong population growth, a 61 percent increase would be required.
AI is enabling more creative and effective agriculture practices, which is greatly increasing crop yield. A study that was published in the National Library of Medicine claims that AI and Internet of Things (IoT)-powered intelligent irrigation systems guarantee accurate and ideal watering, which leads to more water savings and healthier crops.
In agricultural monitoring and breeding, machine learning algorithms provide excellent accuracy in identifying features that result in more resilient plants and greater yields. The study also describes how the technologies facilitate the prompt detection of plant illnesses, occasionally attaining detection accuracy rates of over 95%, which lowers crop loss and boosts output.
Brian Lutz, Vice President of Agricultural Solutions at Corteva Agriscience, spoke with Editorial Director Matthew DeMello recently about how incorporating various AI models—particularly those that analyze genetic and protein data—is revolutionizing Corteva’s biological discovery and product development.
Their discussion demonstrates that AI is not theoretical in the life sciences. It marks a transition from experimental to demonstrated impact by producing genuine, validated discoveries more quickly and affordably than traditional R&D processes.
Two practical takeaways from their discussion are deconstructed in this article:
Breaking silos to speed up discovery: Using a variety of chemical, academic, patent, and agricultural data to navigate large chemical spaces in order to find new, high-impact chemicals.
Combining AI capabilities to accelerate innovation in life science: combining various AI algorithms to enable quicker and more efficient protein and compound discovery.
Combining transformer, diffusion, and domain-specific models to enable quicker, more precise, and scalable innovation in the life sciences is an example of integrating complementary AI models to promote real-world impact.
The complete audio can be heard below:
Corteva Agriscience’s Vice President of Agricultural Solutions, Brian Lutz, is the guest.
Knowledge: Business development, strategy, and leadership
Brief Recognition: At Corteva Agriscience, Brian is responsible for spearheading the creation of innovative goods and services. He is in charge of managing resource investments, finances, and strategic direction for a sizable and heterogeneous team of technological specialists. He graduated with a BA in biology from the College of Wooster and a PhD in biogeochemistry from Duke University.
Dismantling Silos To Hasten Discovery
Brain introduces the topic by saying that the true worth of AI is found in the effect and novelty of what may be learned and accomplished. He admits that AI is changing many facets of business, including operational effectiveness and consumer interactions. He does, however, stress that discovery in the life sciences is essentially distinct from merely streamlining processes.
He uses the estimated number of sand grains on Earth (10^18) as an example, and the estimated number of physiologically significant compounds that may be used to make medications or crop protection chemicals (10^60). He claims that these size extremes show the life sciences’ enormous potential for discovery.
Finding safe, efficient compounds in an extraordinarily large chemical space is the challenge, he says.
Because these models are generalizable, Brian adds that one of the fascinating features of AI in discovery is its capacity to dismantle organizational walls across businesses or industry sectors. He cites Corteva’s work on the Atlas platform with Beloit partners as an excellent illustration.
Much though the platform was first developed for the pharmaceutical sector, combining vast amounts of chemical data makes it much more potent:
The truth is that you can unlock whole new discovery potential when you start creating knowledge graphs across several data modalities and combine vast amounts of chemical data from academic and patent publications.
We can examine the larger chemical space and find novel chemicals that we probably wouldn’t have otherwise discovered when we integrate our agriculture data with that platform. We are therefore going beyond concentrating on a small number of distinct use cases. We can now speed up discovery for our company by combining data from the whole body of chemical knowledge.
— Brian Lutz, Corteva Agriscience’s Vice President of Agricultural Solutions
Using Complementary AI Models To Promote Impact In the Real World
Although AI has advanced significantly from conventional analytical methods to generative and transformer-based models, Brian asserts that there isn’t a single panacea. He stresses that integrating all of these qualities is where the true effect lies. He emphasizes the benefits of broad language models, particularly those made for natural language processing, which allow for human language interaction, reasoning, and idea iteration.
However, Corteva has a crucial use case in a different area of the LLM space: employing transformer models to process DNA sequences and model protein structures rather than words. Brian cites the example of AlphaFold, which won the Nobel Prize for its influence.
For Corteva to optimize proteins in seeds or find compounds that interact with those proteins to preserve crops, an understanding of protein structures is crucial:
In the past, it could take months and cost tens of thousands of dollars, or more, to determine the structure of a single protein in the lab. These days, it just takes a few hours to predict the structure of every protein in a whole genome. It merely takes a few seconds and costs pennies to simulate individual proteins.
Diffusion models and other analytical techniques can be used to examine large chemical libraries and determine which compounds are most likely to interact with those proteins in the ways that we require once we know those structures. These exchanges serve as the foundation for our upcoming product line.
— Brian Lutz, Corteva Agriscience’s Vice President of Agricultural Solutions
According to Brian, the new generation of big transformer models is based on complexity theory, which is frequently disregarded. He claims that these models are used in natural language processing to extract embedded thoughts from word strings.
Beyond that, though, users may now arrange several concepts into intricate debates and even create whole original stories, like creating a book.
He makes a comparison between these same fundamental ideas and how they are used in chemistry and biology. The models, for example, can be used to assess how genes are arranged inside chromosomes, similar to how chapters are arranged in a book, and to differentiate between genes and intergenic regions when applied to genomes. The models thereby convert biological function into the “language” of biology.
Brian highlights that these are emergent patterns, similar to textual thoughts, made possible by the models’ capacity to comprehend complexity and relationships rather than just specific context. He claims that this is how complexity theory works.